Digital image forgery detection using 3d lighting fingerprints

Abstract

With the growing availability and acceptability of digital media in various domains, the risk of image tampering is ever increasing. This increases the need to identify such tampering for authentication of digital media in domains such as jurisdiction, copyright and identity management, content analysis etc. One of the important facts related to this field is that we need to have high confidence in the forgery detection algorithms and approaches. These techniques must be validated and accepted by scientific community based on standard scientific bases. newlineThe motivation of this research started with the assessment of the possible light source direction from an image under single light source. However, this research work further extended to find image tampering using multiple light sources. Very few works exist in the literature which uses complex multiple lighting based forgery detection. In this research work best-in-class forgery detection techniques are proposed which use inconsistencies in the light source directions to detect image forgery. To preprocess the input forged image, surface normals are estimated from the structural profile. The enhancement of the structural and illumination profiles are done by implementing different image enhancement filters. Surface normals are obtained on image object edges for chosen patches instead of using complex shading and shadows which sometimes lead to inaccurate estimation of surface normals. With estimated radiance and surface normals, the angle with respect to light source directions are computed for fake and real objects present in the image. The inconsistency in obtained results show the evidence of tampering based on inherent lighting fingerprints. For more than one light source environments, the elevation angles are computed with respect to each light source present in the scene. Different intensity light sources are used to create an image with composite lighting environment and tested with the proposed techniques. Further, the obtained results are being compar